Ace. Chem. Res. 1995,28, 251-256
251
Determination of RNA Conformation by Nuclear Magnetic Resonance PETER B. MOORE Department of Chemistry, Yale University, New Haven, Connecticut 06520 Received November 2, 1994
Introduction
A
Spectroscopic determination of RNA conformation has recently emerged as a distinct subdiscipline of biophysical chemistry, and as befits its youth, it is in flux. In the past, RNA spectroscopy depended on lH and 31Pexperiments. In the future, experiments that use samples labeled with 13C and 15N will dominate. This account comments on both approaches and on the nature of the information they provide.
00”
-
A5
OH
bH
bH
Historical Background Although it has been recognized for decades that the function of many RNAs depends on their conformations, it is still true that only a handful of RNA structures are known at atomic resolution. The reason is that, until the late 1980s, no RNAs except those available in nature could be characterized structurally, and only a few can be obtained that way in the amounts required: the transfer RNAs (tRNAs), the ribosomal RNAs, and viral RNAs. With the exception of the 5s RNA from ribosomes (Mr = 40 000), only the tRNAs had molecular weights small enough t o appear workable (Mr = 25 000). The systematic application of nuclear magnetic resonance (NMR) to RNA began in 1971 when it was discovered that resonances due t o hydrogen-bonded imino protons in RNA base pairs, GN1 and UN3 protons (Figure 11, can be detected between 10 and 15 ppm, downfield of all other RNA proton resonances.lJ Imino proton spectra tend to be wellresolved. Those given by tRNAs, for example, contain roughly 25 resonances, about one per base pair, and they are dispersed over a spectral region as wide as that which contains the molecule’s 600 nonexchangeable proton resonances (see Figure 3). Furthermore, downfield spectra are worth studying because assigning an RNA’s downfield spectrum is approximately equivalent t o determining how it is base paired. Pure tRNAs were available in 1971, and tRNA conformation was not understood. Thus, provided resonances could be assigned, a significant problem could be solved. By the time it was discovered that imino proton spectra can be assigned by NOE spect r o s ~ o p y , ~however, -~ the tRNA problem had been solved crystallographically. Peter B. Moore, who was born in Boston, MA, in 1939, got his B.S. from Yale University in 1961 and his Ph.D. from Harvard University in 1966, both in biophysics. After three years as a postdoctoral fellow at the University of Geneva (Institut de Biologie Moleculaire) in Switzerland and the Medical Research Council Laboratory of Molecular Biology, Cambridge (U.K.), he joined the faculty of the Department of Molecular Biophysics and Biochemistry at Yale. He is now professor of chemistry at Yale. with a joint appointment in molecular biophysics and biochemistry, The focus of his research is the structural characterization of the RNAs and ribonucleoproteins involved in protein synthesis.
0001-4842/95/0128-0251$09.00/0
0
C
D
II
I
C
H.c/
NN
1I
I
H Oc\N/c\o
I
H
Figure 1. Structures of the RNA nucleotides: (A) adenosine 5’-monophosphate, with the number conventions indicated for its purine and ribose rings; (B) uracil; (C) guanine; (D)cytosine. The other 5’ monophosphates are obtained by replacing the purine ring of adenosine with uracil, guanine, and cytosine, respectively. Purines are linked to riboses by Cl’-N9 glycosidic bonds. The linkage for pyrimidines is Cl’-Nl. B
A R
c + s e
51
o=p-o-
I R’
H
H 4’ H H 3’ 3’
Figure 2. Nucleotide conformation: (A) The torsion angles associated with each nucleotide are indicated. R and R‘ are the nucleotides on the 5’ and 3’ sides of the one shown, respectively. “Base”is any one of the four heterocyclic bases shown in Figure 1.(B) The conformations of ribose when its puckering is the C3’endo (north) (top) and C2’-endo (south) (bottom) are shown.
Emphasis shifted from imino proton spectroscopy to full conformational characterization in the late 1980s because of two developments: (1) the discovery of chemicallo and enzymaticll methods for synthesizing RNAs, which liberated the field from its preparative (1)Kearns, D. R.; Patel, D.; Shulman, R. G. Nature 1971,229, 338339.
0 1995 American Chemical Society
252 Ace. Chem. Res., Vol. 28, No. 6, 1995
Moore Amino and aromalic protons
Imino protons
PPm
8.5
8.0
7.5
7.0
6.5
6.0
5.5
5.0
4.5
4.0
wm Figure 3. Typical RNA proton spectra. Top: the downfield part of an RNA spectrum taken in H20. Bottom: an entire RNA spectrum taken in DzO. The sharp resonance a t 4.7 ppm is residual HDO, and the sharp resonances at about 3.8 ppm are buffer components. The types of resonances expected in each region are indicated. (Reprinted with permission from the Ph.D. thesis of A. Szewczak, Yale University, 1994. Copyright 1994, A. Szewczak.)
bind, and (2)the evolution of methods for determining macromolecular conformations by NMR.12 The discovery of a host of new, stable RNAs t o study in the 1970s and 1980s and the discovery of RNA catalysis13J4contributed mightily t o the field’s vigor.
The Problem To Be Solved Four nucleotide monomers predominate in RNA: two purines, A (adenosine 5’-monophosphate) and G (guanosine 5’-monophosphate), and two pyrimidines, U (uridine 5’-monophosphate) and C (cytidine 5’monophosphate) (Figure 1). (Modified nucleotides do occur in nature, but so far, none of the RNAs characterized in detail spectroscopically has included any.) Adjacent monomers are linked 5‘ t o 3’ by phosphodiester bonds. RNAs adopt folded conformations, which are stabilized by base stacking and by hydrogen bonding between the donors and acceptors, which abound in RNAs. Their conformations can be described by specifying seven torsion angles per nucleotide monomer (Figure 2). Six of the seven describe the trajectory of the molecule’s backbone, and the seventh, x, specifies the orientation of bases relative to ribose rings. The anti conformation, in which purine N l s and pyrimidine N3s point away from ribose rings, predominates in RNA, but occasional residues are found in the syn conformation, where purine N l s and pyrimidine N3s lie over their ribose rings. The ribose (2) Kearns, D. R.; Patel, D.: Shulman, R. G.; Yamane, T. J . Mol. Biol. 1971, 61, 265-270. ( 3 )Johnston, P. D.; Redfield, A. G. Nucl. Acids Res. 1978, 4, 35993615. (4)Johnston, P. D.; Redfield, A. G. Biochemistry 1981,20,1147-1156. (5) Sanchez, V.; Redfield, A. G.; Johnston, P. D.; Tropp, J. Proc. Nut. Acad. Sci. U.S.A. 1980, 77, 5659-5662. ( 6 )Roy, S.; Redfield, A. G. Nucleic Acids Res. 1981, 9 , 7073-7083. ( 7 )Hare, D. R.; Reid, B. R. Biochemistry 1982, 21, 1835-1842. ( 8 ) Hare, D. R.; Reid, B. R. Biochemistry 1982, 21, 5129-5135. (9) Roy, S.;Papastarvos, M. Z.; Redfield, A. G. Biochemistry 1982,
____
27_ 6, n R l - f “ _
(10)Usman, N.; Ogilvie, K. K.; Jiang, M.-Y.; Cedergren, R. J. J . A m . Chem. SOC.1987, 109, 7845-7854. (11) Milligan, J. F.; Groebe, D. R.; Wetherell, G. W.; Uhlenbeck, 0. C. Nucleic Acids Res. 1987, 15, 8783-8798. (12) Wuthrich, K. NMR ofProteins and Nucleic Acids; John Wiley & Sons: New York. 1986 (131 Kruger, K.; Grabowski, P. J.; Zaug, A. J.; Sands, J.;Gottschling, D. E.: Cech, T. R. Cell 1982, 31, 147-157. (14) Guerrier-Takada, C.; Gardiner, K.; Marsh, T.; Pace, N.; Altman, S. Cell 1983, 35, 849-857.
ring puckers that predominate are C3’-endo (abundant) and C2’-endo (rare)15(Figure 2). Sugar pucker fixes one of the backbone torsion angles, 6, as well as all the other ribose torsion angles. In principle, the most efficient way t o work out an RNA’s conformation would be t o determine its torsion angles directly: by measuring three-bond coupling constants, for example. But, even if all the relevant couplings could be measured, unrealistically accurate torsion angle estimates would have to be derived from them t o obtain reasonable structures. In practice, heavy reliance is placed on proton-proton distances derived from nuclear Overhauser effect (NOE) intensities. Whatever the mixture of coupling constants and NOES used, as Figure 3 shows, the spectroscopic information required for conformation determination must be extracted from spectra that are intrinsically poorly resolved.
Molecular Weight and the Choice of RNAs for Study For any class of macromolecules there is always a molecular weight beyond which the techniques of the day cannot assign spectra because resonances are too numerous and too broad. The limit is in the neighborhood of 10 000 (30 nucleotides) for RNA today, but the new heteronuclear approaches discussed below, combined with increases in spectrometer field strength, could double it in the next few years. Unfortunately, almost no naturally occurring RNAs have chain lengths as short as 60 nucleotides. However, the secondary structure of the typical RNA is dominated by hairpin-like structures that consist of an intramolecular helical stem with a loop that closes its distal end, and its tertiary structure is generated by interactions between these stendloops and between them and less ordered sequences.16 Many sterdoops are small enough to analyze spectroscopically, and small RNA motifs like these often retain their conformations in isolation. While Watson-Crick pairing (A with U, and C with G ) dominates in small RNA motifs, they often include other juxtapositions and bases that lack obvious (15)Altona, C. Rec: J . R . Neth. Chem. Soc. 1982, 101, 413-433 116) Gutell, R. R. Nucl. Acids Res. 1993, 21, 3051.
Determination of RNA Conformation hydrogen-bondingpartners.17 Thus the conformations of many small RNAs cannot be inferred from sequence alone. Furthermore, many RNA binding proteins interact with single stem/loops, and catalytically active RNAs below the molecular weight limit can sometimes be prepared by trimming natural sequences. There is a lot to be learned from the study of the conformations of RNA oligonucleotides. Unfortunately, RNAs cannot be selected for study on the basis of size and interest alone. Many sequences aggregate a t NMR concentrations (L 1 mM), and others lack unique structures. While these difficulties can sometimes be palliated by changes in temperature and/or ionic conditions, most sequences that show either propensity cannot be analyzed spectroscopically. About 75% of the RNA investigations we undertake are abandoned for these reasons. Once a suitable RNA is identified, the next step is to assign its proton spectrum.
Acc. Chem. Res., Vol. 28, No. 6, 1995 253 “Traditional” ‘H-driven methods for assigning RNA spectra start with base resonances and work in toward the backbone, i.e., from the “outside”of the molecule, covalently speaking, to its “inside”. They are largely NOE-driven. The new heteronuclear methods work from the backbone out t o the bases and depend on scalar couplings (see below). The imino proton assignment method just described, which is a fallible (!), bases-in method, is the only one available today. A more robust, heteronuclear method, which depended on correlations between imino proton and nonexchangeable proton resonances generated by scalar couplings, would be welcome indeed. Why Nonexchangeable Proton Resonances Are Hard To Assign
The assignment of RNA proton resonances is challenging for several reasons. First, base protons are not scalar coupled to ribose protons, and the protons within a single ribose (usually) do not constitute a Imino Proton Assignments single, scalar-coupled spin system either.25 Consequently, proton total correlation spectroscopy (TOCSY) Imino proton resonances in RNA helices, which are and correlated spectroscopy (COSY) experiments, the always A-form,17 are (usually) easy t o assign because through-bond experiments protein spectroscopists use the imino protons in successive base pairs are close t o identify resonances belonging t o single residues,12 enough to give through-space, NOE correlations.6 are much less useful for RNA. Second, while it is easy Thus every helical segment in an RNA engenders a t o distinguish purines from pyrimidines (pyrimidine set of NOE-connectableimino proton resonances. Less H6 protons are coupled to H5 protons, and purine H8 regularly structured regions often do the same. protons are not coupled t o anything) there is no easy UN3 resonances in Watson-Crick AU pairs resot o distinguish Us from Cs and Gs from As. Third, way nate on the downfield side of the imino proton region the backbone of an RNA is the only part of the (13-15 ppm) and give intense NOEs t o the (narrow) molecule where covalent chemistry constrains atoms AH2 proton resonances of their hydrogen-bonding to relate to each other in a spectroscopically benign partner^.^ GN1 protons in Watson-Crick GCs, which way. The backbone-associated protons of an RNA, usually resonate between 12 and 13 ppm, give strong however, are its H3’, H4‘, H5’, and H5” protons, which NOEs to the two (exchangeable) CN4 protons of their resonate around 4 ppm, where resonance overlap is hydrogen-bonding partners and weaker, transferred its worst. Thus the resonances most likely to yield a t NOEs t o CH5s.12 GUS, the most common non-Watconformation-independent sequential assignment inson-Crick pairing in RNA, are also easy to i d e n t i f ~ . ~ to observe. formation are the hardest Once imino proton resonances have been sorted by base pair type and ordered on the basis of iminoThe Bases-In Strategy imino NOEs, sequences likely t o form stemsls can be correlated with imino proton runs and assignments The bases-in strategy depends on anomeric (H1’)can be made. aromatic (H6, H8) NOEs, which are found in one of Imino proton runs in regions that are irregularly the better resolved regions of a n RNA’s nuclear paired are matched with sequences using base type Overhauser effect spectroscopy (NOESY) specas the criterion, instead of base-pair type. The base t r ~ m . ~The ~ ,H1’ ~ (anomeric) ~ - ~ ~ proton of every ribose type of an imino proton resonance can be determined is (always) within NOE range of its own aromatic (H8 by experiments that correlate imino proton resonances or H6) proton. In addition, in helical stems, the H1’ with the 15Nresonances of the nitrogens to which they of residue n is within NOE range of the H6 or H8 of are bonded. These correlations are best observed residue (n 1). Thus every anomeric proton in a using 15N-labeled ~ a m p l e s , l but ~ - ~also ~ can be seen stem, except the one a t its 3’ terminus, gives NOE using the 15Npresent in unlabeled samples.23 In any cross peaks t o its own (H6, H8) and that of the next case, these experiments work because the 15Nchemical base in the sequence, and every (H6, H8) except the shift ranges of UN3s and GNls do not overlap.24 one a t the 5’ terminus of a sequence gives NOEs to two H1’ proton resonances. The result is that the (17) Saenger, W. Principles of Nucleic Acid Structure; Springeraromatic-anomeric part of a helical RNA’s NOESY Verlag: New York, 1984. spectrum includes rectangular patterns of cross peaks (18)Turner, D.; Sugimoto, N.; Freier, S. M. Annu. Rev. Biophys.
+
Biophys. Chem. 1988, 17, 167-192. (191 Bax, A,; Griffey, R. H.; Hawkins, B. L. J . Magn. Reson. 1983,55, 301-335. ~
~
~~~
(201 Griffey, R. H.; Poulter, C. D.; Bax, A,; Hawkins, B. L.; Yamaizumi, Z.; Nishimura, S. Proc. Natl. Acad. Sci. U.S.A. 1983, 80, 5895-5897. (21) Griffey, R. H.; Redfield, A. G.; Loomis, R. E.; Dahlquist, F. W. Biochemistry 1985, 24, 817-822. (22) Kime, M. J. FEBS Lett. 1984, 175, 259-262. (231 Szewczak, A. A.: Kellogg, G. W.; Moore, P. B. FEBS Lett. 1993, 327, 261-264. (241 Gonnella, N. C.; Birdseye, T. R.; Nee, M.; Roberts, J. D. Proc. Natl. Acad. Sci. U.S.A. 1982, 79, 4834-4837.
(25) de Leeuw, F. A. A. M.; Altona. C. J. Chem. SOC..Perkin Trans. 2 1982, 375-384.
(26) Feigon, J.; Leupin, W.; Denny, W. A.; Kearns, D. R. Biochemistry
27 , 5942-5‘Xl -1BR.1 - -- , __ - - -- -- - -.
(27) Hare, D. R.; Wemmer, D. E.; Chou, S.-H.; Drobny, G.; Reid, B. R. J . Mol. B i d . 1983, 171, 319-336. (281 Haasnoot, C. A. G.; Heershap, A,; Hilbers, C. W. J . Am. Chem. Soc. 1983. 54RR-54R4. - ~~ . . , -105. _ .., _ - ..- . .
(29) Scheek, R. M.; Russo, N.; Boelens, R.; Kaptein, R.; van Boom, I. H. J. Am. Chem. Sac. 1983, 105, 2914-2916. (301 Varani, G.; Tinoco, I., J r . Q.Reu. Biophys. 1991, 24, 479-532.
Moore
254 Acc. Chem. Res., Vol. 28, No. 6, 1995
called “anomeric-aromatic walks” that can be used for making sequential assignments. Helix-like, anomeric-aromatic walks often occur in irregularly paired regions as well. Sometimes anomeric-aromatic walks can be assigned to sequences by the ordering of purine and pyrimidine bases implied. More often, anomericaromatic walks are placed by correlating them with assigned imino proton resonances. In helical regions, the AH2 proton of each AU base pair, which is identified by its UN3-AH2 NOE, gives NOES to the H1’ protons of the nucleotides on the 3’ side of both the A and the U.12 Similarly, a t long mixing times, NOES can be seen between GN1 imino protons in GCs and the H1’ protons of the residues on the 3’ sides of both the G and the C.31 Once H1’ assignments have been made, other ribose resonances can be addressed. Most H2’ resonances are assignable because H2’-H1’ cross peaks dominate the anomeric-ribose region of RNA NOESY spectra, and because H2’-aromatic walks exist analogous to the anomeric-aromatic walks just discussed. H3’ assignments are harder t o make because most H2’H3’ NOES fall in the ribose-ribose region. They can usually be pieced together, however, from H3’aromatic walks and from 31P-1H coupling information (see below). Bases-in assignment tends to stall a t H3’ unless the RNA being examined is so small that its ribose-ribose region is resolvable. Complete proton assignments have been obtained in a few such cases, and even H5’s distinguished from H ~ ’ ’ s . ~ ~ TOCSY and COSY experiments contribute relatively little because Hl’-H2’ couplings are very small in ribose rings with C3’-endo puckeringeZ5TOCSY data will correlate the H2’, H3’, and H4‘ resonances of a C3’-endo ribose, but these correlations are often hard to identify because they fall in the ribose-ribose region. A C2‘-endo puckered ribose gives useful TOCSY correlations, but only Hl’, H2’, and H3’ chemical shifts are correlated because H3’-H4’ couplings are small. Hl’-H2’-H3’-H4‘ TOCSY correlations are seen only when sugars are exchanging rapidly between the two puckers. H4’-(H5’,H5”) coupling constants can be appreciable, but their magnitudes are strongly affected by backbone torsion angles and thus cannot be relied upon for assignments. The risk of assignment error can be minimized by experiments that identifjr resonances by chemical type. AH2s can be recognized using inversion-recovery experiments because their 2’1s are much longer than those of other aromatic protons. Natural abundance 13C-lH correlation experiments are very useful also.34 Since the 13Cchemical shifts of RNA carbon atoms fall into (largely) nonoverlapping ranges determined by chemical type, proton resonances can be sorted by chemical type on the basis of the chemical shifts of the 13Csto which they are bonded. Selective deuteration can be used t o sort H6 and H8 resonances by base type. d5-U, d5-C, d8-A, and d8-G are easy to prepare (31)Heus, H.; Pardi, A. J . Mol. Biol. 1991,217, 113-124. (32) Pardi, A,; Walker, R.; Rapoport, H.; Wider, G.; Wuthrich, K. J . A m . Chem. SOC. 1983,105, 1652-1653. (33)See: Hines, J.;Varani, G.; Landry, S. M.; Tinoco, I., J r . J . A m . Chem. SOC. 1993,115, 11002-11003 and references therein. (34)Varani, G., Tinoco, I., J r . J . A m . Chem. SOC. 1991,113, 93499354.
by exchange starting with protonated nucleosides (or nucleotide^)^^-^^ and to convert into RNA (see below). The weaknesses of the bases-in approach are obvious. It seldom yields a complete set of assignments, and it is conformation-dependent. It can fail completely when the conformation of some part of an RNA does not lead to helix-like aromatic-anomeric NOE connectivities or, worse yet, leads t o something that looks like the helical pattern but is not. Unhappily, the regions where assignment is difficult are always the most interesting.
31P-1H Spectroscopy: A Step toward Backbone-Out Assignments Since the predominant isotope of phosphorus, 31P, has a nuclear spin of l/2, all RNAs have a 31Pspectrum that contains (about) one resonance per residue. Assignment of an RNA’s 31P spectrum is useful because 31Pchemical shifts are sensitive to a and 5,38 and so 31P resonances with unusual chemical shifts are associated with regions having unusual conformations. More important, every 31P atom is scalar coupled t o the 3’ proton of the ribose on the 5’ side of its phosphate group, and t o the 5’ and/or 5” protons of the ribose on the 3’ side of its phosphate. (3’ side 31P/H4‘couplings are also sometimes large enough t o observe.) The first backbone-out, sequential assignment strategy proposed for nucleic acids was based on these couplings, but required that cross peaks be resolved in the ribose-ribose region, and so was applicable only to small RNAs.32 A much less restrictive approach became possible when 31P/1Hhetero-TOCSY experiments were developed by K e l l ~ g g . ~These ~-~~ experiments can generate 31P-Hl’ and sequential 31Paromatic correlations. At worst, they allow one to assign 31P resonances on the basis of anomeric and aromatic correlations. At best, something approaching a complete set of 31Pand IH assignments for a small RNA or DNA can emerge from a single e ~ p e r i m e n t . ~ ~ , ~ ~ 31P-1H experiments are not the final answer, however. The three-bond couplings they exploit depend on c ~ n f o r m a t i o nand , ~ ~ the clearest sequential walks they generate in RNA involve phosphate-aromatic correlations that depend on interresidue NOES. Finally, the dispersion in the 31Pdimension of 31P-1H spectra tends to be poor because the prevalence of A-form torsion angles in RNA leads to resonance overlap.
The Heteronuclear, Backbone-Out Revolution Starting in about 1990, scalar coupling methods for assigning protein spectra were developed that use samples uniformly labeled with 13C and 15N.42-44 (35) Benevides, J. M.; Lemeur, D.; Thomas, G. J., J r . Biopolymers 1984,23,1011-1024. (36) Brusch, C . K.; Stone, M. K.; Harris, T. M. Biochemistry 1988, 27, 115-122. (37) Hayatsu, H.Prog. Nucleic Acid Res. Mol. Biol. 1976,16, 75124. (38) Gorenstein, D. G., Ed. Phosphorus-31 NMR; Academic Press, Inc.: New York, 1984. (39) Kellogg, G. W. J . Mugn. Reson. 1992,98, 176-182. (40) Kellogg, G. W.; Szewczak, A. A,; Moore, P. B. J . A m . Chem. Soc. 1992,114, 2727-2728. (41)Kellogg, G. K.; Schweitzer, B. I. J . Biomol. NMR 1993,3 , 577595. (42) Bax, A,; Clore, G. M.; Gronenborn, A. M. J . Magn. Reson. 1990, 88, 425-431.
Determination of RNA Conformation Resolution is routinely improved in these experiments by dispersing proton-proton cross peaks using associated heteroatom chemical shifts. In addition, assignments are conformation-independent because the oneand two-bond couplings these experiments exploit are largely independent of torsion angles. The RNA community was not slow t o grasp the potential of this approach. It is easy t o prepare 13C- and/or 15N-labeled RNA samples. Bacterial cells are grown on a minimal medium in which 15N NH4’ salts are the sole source of nitrogen and/or 13C glucose45or 13C methanol46is the sole source of carbon. The RNA recovered from these cells is digested to 5’ nucleoside monophosphates with RNAse P1 and then enzymatically converted into nucleoside triphosphates or nucleosides, the substrates for enzymatic and chemical RNA synthesis, respe~tively.~~J~~~~~~~ A 13C-15N-labeled nucleotide is a single-spin system. Its ribose protons can be completely assigned using HCCH experiment^.^^ Experiments of the HCN and HCNCH class will correlate Hl’s with H8s and H6s,j0-j3 and adenine H8s can be correlated with adenine H2s in much the same way.54j55 Finally, experiments of the HCP class can generate conformation-independent, through-bond, backbone-based, sequential assignment^.^^-^* 13C and 15N assignments are produced almost as a byproduct of this approach, and once those resonances are assigned, 13C-13C and 13C-lH couplings can be used to obtain torsion angles. Many of the heteronuclear experiments just alluded to involve pulse trains that consume times that are They are bound to significant compared t o RNA TZS. be less effective for RNAs larger than the ones on which they have been demonstrated because of signal loss resulting from reduced transverse relaxation times. Heteronuclear labeling aggravates this problem by opening additional relaxation pathways in macromolecules. We find, for example, that 13C-edited NOESY experiments done on specifically labeled RNAs have noticeably poorer signal to noise than unedited experiments done on comparable unlabeled samples (Kellogg, G. W., and Moore, P. B., unpub(43) Bax, A,; Clore, G. M.; Driscoll, P. C.; Gronenborn, A. M.; Ikura, M.; Kay, L. E. J . Mugn. Reson. 1990, 87, 620-627. (44) Fesik, S. W.; Eaton, H. L.; Olejniczak, E. T.; Zuiderweg, E. R. P. J . Am. Chem. SOC.1990, 112, 886-888. (45) Nikonowicz, E. P.; Sirr, A,; Legault, P.; Jucker, F. M.; Baer, L. M.; Pardi, A. Nucleic Acids Res. 1992,20, 4507-4513. (46) Batey, R. T.; Inada, M.; Kujawinski, E.; Puglisi, J. D.; Williamson, J. R. Nucleic Acids Res. 1992, 20, 4515-4523. (47) Haynie, S. L.; Whitesides, G. M. Appl. Biochem. Biotechnol. 1990, 23, 205-220. (48) Batey, R. T.; Battiste, J. L.; Williamson, J. R. Methods Enzymol., in press. (49) Pardi, A,; Nikonowicz, E. P. J . Am. Chem. SOC.1992,114,92029203. (50) Farmer, B. T., 11; Muller, L.; Nikonowicz, E. P.; Pardi, A. J. Am. Chem. SOC.1993, 115, 11040-11041. (51)Sklenar, V.; Peterson, R. D.; Rejante, M. R.; Feigon, J. J. Biomol. NMR 1 9 9 3 , 3 , 721-727. (52) Sklenar, V.; Peterson, R. D.; Rejante, M. R.; Wang, E.; Feigon, J. J . Am. Chem. SOC.1993, 115, 12181-12182. (53) Farmer, B. T., 11; Muller, L.; Nikonowicz, E. P.; Pardi, A. J . Biomol. NMR 1 9 9 4 , 4 , 129-133. (64) Marino, J. P.; Prestegard, J. H.; Crothers, D. M. J . Am. Chem. SOC.1994, 116, 2205-2206. (55) Legault, P.; Farmer, B. T., 11; Mueller, L.; Pardi, A. J . Am. Chem. SOC.1994, 116, 2203-2204. (56) Marino, J. P.; Schwalbe, H.; Anklin, C.; Bermel, B.; Crothers, D. M.; Griesinger, C. J . Am. Chem. SOC.1994, 116, 6472-6473. (57) Marino, J. P.; Schwalbe, H.; Anklin, C.; Bermel, W.; Crothers, D. M.; Griesinger, C. J . Biomol. NMR, in press. (58) Heus, H. A,; Wijmenga, S.S.; van de Ven, F. J. M.; Hilbers, C. W. J . Am. Chem. Soc. 1994, 116, 4983-4984.
Acc. Chem. Res., Vol. 28, No. 6, 1995 255 lished observations). Also proton chemical shifts have an irritating tendency to correlate with 13C chemical shifts, which reduces the dispersion achieved by labeling IH,lH cross peaks with I3C frequencies. In the end, heteronuclear methods may contribute more t o the field by making complete assignment of small RNAs possible than by increasing the molecular weight limit.
Diagnosis of RNA Topology Once proton resonances have been assigned, two of each nucleotide’s seven torsion angles, x and 6 , can be determined. Nucleotides can be classified as anti or syn because intranucleotide H1’-aromatic NOEs are weak in the former case and strong in the latter, and ribose ring puckers can be identified as C3’ endo or C2’ endo (or some mixture of the two) on the basis of Hl’-H2’ couplings. The pairings of imino proton carrying bases in irregular regions can also be diagnosed using iminoother NOEs. Seventeen of the 20 possible base pairings involving Gs and Us put hydrogen-bonded imino protons within NOE distance of protons from partner bases.17 The other three sometimes give diagnostic imino-other NOEs as well.59 When more than one pairing is compatible with an imino-other NOE(s), the decision can often be made on the basis of glycosidic torsion angle. Opposing strands in a double helix or a closed loop are necessarily antiparallel, and in antiparallel structures, the ribose of one base cannot be superimposed on the ribose of its partner by rotation about an axis perpendicular to the plane of the pair. Thus pairings that are rotationally symmetric are unlikely unless one of the participants is in the syn conformation. (Exceptions have been f o ~ n d ! ~Remaining ~ , ~ ~ ) problems can usually be resolved by model building; one option will fit, and the other(s) will not.
Determination of RNA Conformations Distance geometry algorithms provide the best route for proceeding from NMR data to macromolecular conformations (see the recent review of Brunger and NilgesG1). The input includes the sequence of the RNA, a list of standard bond lengths and angles, a description of the molecule’s base-pairing pattern, NOE-derived distances, and coupling-constant-derived torsion angles. The data are converted into information about the range of distances possible between all pairs of atoms in the molecule, and models are generated consistent with those ranges, which are then refined. Refinement invariably involves the minimization of a structure’s nominal energy as well as reduction of its deviation from the experimental data. Energy minimization is essential t o ensure proper bond lengths and angles, and to prevent close contacts, but it does mean that the model which emerges is “invented” by the minimization program’s energy functions t o some degree. This would be acceptable if the potentials used were entirely accurate, but they are (59) Szewczak, A. A.; Moore, P. B. J . Mol. Biol. 1995, 247, 81-98. ( 6 0 )Wimberly, B.; Varani, G.; Tinoco, I., Jr. Biochemistry 1993, 32, 1078-1087. (61) Brunger, A. T.; Nilges, M. Q.Reu. Biophys. 1993, 26, 49-125.
Moore
256 Ace. Chem. Res., Vol. 28, No. 6, 1995
not. They seldom (never?) treat electrostatic interactions rigorously, for example. What impact does energy minimization have on the structures that emerge? No general answer can be given: (1)the amount of experimental data available varies from one structure determination t o the next; (2) error propagation is an inexact art in the NMR world, t o put it mildly; and (3) there are no conventions about how contributions t o molecular energy should be weighted during refinement relative either t o experimental data or to each other. It follows that the superpositions of independently computed structures, which are used to illustrate the reliability of structures, cannot be compared from one paper t o the next because the variation depicted depends on the variables just mentioned. These issues are particularly sensitive ones for RNA spectroscopists because the data they obtain seldom determine their structures uniquely.63 In our experience, 15-20 experimental constraints per nucleotide can be obtained from fully assigned RNA’s proton spectra;62one often must settle for less. This is more than the minimum required (seven), but the experimental constraints are poorly distributed. They usually determine base placements well, but leave many backbone torsion angles virtually undetermined. In addition, inter-residue constraints that speak t o the relative placement of residues that are unrelated by secondary structure are vanishingly few in number. One symptom of the shortage of information is the high failure rate that accompanies the derivation of RNA models by distance geometry methods. A “failure” is a model that cannot be refined so that it has good covalent geometry, low energy, and good agreement with the NMR data. (Fortunately, in our experience, good geometry, low energy, and good agreement with the data have always gone hand-inhand.) Failure occurs because the information supplied is insufficient to prevent the construction of models that include conformational discrepancies so serious that they cannot be refined away. Failure rates of 80-90% are common. In many cases, what should be sought is not so much an accurate representation of an RNA’s conformation as a sensible portrayal of the conformational implications of the NMR data. I feel that one should compute NMR-derived RNA structures that are as similar t o A-form RNA as they can be, except where the data require otherwise, and represent them to the public as being no more than that. When computations are done this way, for example, those backbone torsion angles for which no direct experimental information is available, but which are unlikely to be other than A-form-like, are constrained t o fall within the A-form range, and base pairs are (mildly) restrained to be (62)E.g.: Cheong, C.; Moore, P. B. Biochemistry 1992,31,8406-8414. ( 6 3 ) van de Ven, J. M.; Hilbers, C. W. Eur. J . Biochem 1988,178, 1-38. (64) Heus, H. A.; Pardi, A.Science 1991,253,191-194.
~ 1 a n a r . In l ~ our experience, the failure rate drops, and models result that have the topology required and low energies, that violate few, if any, of the quantitative data, and that look the way crystallographically determined RNAs do. Another way t o state it is that there have been highly A-like conformations within the universe allowed by the data for every molecule we have analyzed. Faut de mieux, it seems natural to regard them as best estimates for the structures in question. It should be emphasized, on the other hand, that families of models calculated in this way give the impression that they are better determined experimentally than is actually the case. Clearly, the field must continue t o search for ways to better constrain its RNA models experimentally. One would like t o be able to arrive a t structures that are so well determined that modest changes in the potentials used during refinement do not matter, and the assumption of A-likeness need not be invoked.
Conclusions While the application of heteronuclear techniques should improve the quality of spectroscopicallydetermined RNA structures, only a few are likely t o be determined with the accuracy protein spectroscopists regularly achieve. That said, it must be recognized that the aspects of a n RNA’s structure that can be determined definitively by NMR, that is, the way its bases are paired and its topology, are the ones that most concern nucleic acid biochemists and molecular biologists. The enterprise is validated by the fact that results of general importance have already begun to emerge. It is becoming increasingly evident, for example, that natural RNAs contain irregularly paired, secondary structure motifs that are as characteristic as the double helix. The sarcidricin loop from ribosomal RNA, which consists of a GNRA tetraloop joined t o a bulged G motif, is a good example.59 Both motifs occur in many other context^.^^,^^ Our capacity t o predict RNA conformations from sequences, which is a major goal of those who investigate RNA structure, will improve as the number of these special secondary structure motifs identified and characterized increases. What the field needs most a t this point, beyond what individuals can achieve by determining more structures and determining them better, are collective decisions on how the errors associated with NMR data should be assessed, on the conventions t o be used when computing models, and on how the accuracy of models should be estimated. Z thank John P. Marino for his help with issues related to heteronuclear spectroscopy. This work was supported by a grant from the National Institutes of Health (GM-41651). AFt940073L
